Building low frequency model with Deep Learning for seismic inversion in complex geology without structural model

Abstract

Summary The conventional low frequency model (LFM) have limitations: uncertainty of spatial variability away from the wells, the uncertainty of the structural model and stratigraphic architecture. It is also challenging to build complex geology structural model. We propose using Deep Feed-forward Neural Network (DFNN) with attributes from seismic partial stacks and seismic velocity to create LFM of elastic properties for Constrained Sparse Spike Inversion. The methodology incorporates training of well curves, additional information from seismic partial stacks and trend from seismic velocity and wells. It has shorter turnaround by not having to include structural model, and is suitable for complex geological settings.